[ 
https://issues.apache.org/jira/browse/BEAM-10265?focusedWorklogId=763721&page=com.atlassian.jira.plugin.system.issuetabpanels:worklog-tabpanel#worklog-763721
 ]

ASF GitHub Bot logged work on BEAM-10265:
-----------------------------------------

                Author: ASF GitHub Bot
            Created on: 28/Apr/22 17:08
            Start Date: 28/Apr/22 17:08
    Worklog Time Spent: 10m 
      Work Description: zhoufek commented on code in PR #17477:
URL: https://github.com/apache/beam/pull/17477#discussion_r860920916


##########
sdks/java/core/src/main/java/org/apache/beam/sdk/schemas/utils/StaticSchemaInference.java:
##########
@@ -87,22 +88,33 @@ enum MethodType {
    * public getter methods, or special annotations on the class.
    */
   public static Schema schemaFromClass(
-      Class<?> clazz, FieldValueTypeSupplier fieldValueTypeSupplier) {
+      Class<?> clazz,
+      FieldValueTypeSupplier fieldValueTypeSupplier,
+      HashSet<Class> alreadyVisitedSchemas) {
+    if (alreadyVisitedSchemas.contains(clazz)) {
+      throw new RuntimeException(
+          "Cannot infer schema with a circular reference. Class: " + 
clazz.getTypeName());
+    }
+    alreadyVisitedSchemas.add(clazz);
     Schema.Builder builder = Schema.builder();
     for (FieldValueTypeInformation type : fieldValueTypeSupplier.get(clazz)) {
-      Schema.FieldType fieldType = fieldFromType(type.getType(), 
fieldValueTypeSupplier);
+      Schema.FieldType fieldType =
+          fieldFromType(type.getType(), fieldValueTypeSupplier, 
alreadyVisitedSchemas);
       if (type.isNullable()) {
         builder.addNullableField(type.getName(), fieldType);
       } else {
         builder.addField(type.getName(), fieldType);
       }
     }
+    alreadyVisitedSchemas.remove(clazz);

Review Comment:
   Looking at this, I'm wondering if we should instead track the type through a 
`Map<Class, Schema>` instead of `Set<Class>`. It might be possible that 
multiple types nested under the same type reference each other in a 
non-circular way. We might be able to avoid repeating work if we keep a 
reference to previously computed schemas. Basically, it would look like:
   
   ```
   if (visited.containsKey(clazz)) {
     Schema value = visited.get(clazz);
     if (value == null) {
       throw new IllegalArgumentException(...);
     }
     return value;
   }
   visited.put(clazz, null);
   // Compute
   visited.replace(clazz, schema);
   ```
   
   Note: I haven't tested this out. I'm just throwing out an idea.





Issue Time Tracking
-------------------

    Worklog Id:     (was: 763721)
    Time Spent: 1h 40m  (was: 1.5h)

> GetterBasedSchemaProvider#schemaFor stack overflows when given a recursive 
> schema
> ---------------------------------------------------------------------------------
>
>                 Key: BEAM-10265
>                 URL: https://issues.apache.org/jira/browse/BEAM-10265
>             Project: Beam
>          Issue Type: Bug
>          Components: sdk-java-core
>    Affects Versions: 2.21.0, 2.22.0
>            Reporter: Reza ardeshir rokni
>            Assignee: Andrei Gurau
>            Priority: P3
>              Labels: Clarified, starter
>          Time Spent: 1h 40m
>  Remaining Estimate: 0h
>
> Proto:
> message TSFoo {
>  string a = 1;
>  string b = 2;
>  TSFoo theOlderMe = 3;
> }
> new ProtoMessageSchema().schemaFor(TypeDescriptor.of(Foo.TSFoo.class));
> Causes a stackoverflow. 
>  
>  



--
This message was sent by Atlassian Jira
(v8.20.7#820007)

Reply via email to